4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
60 #include <linux/kmod.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
67 EXPORT_SYMBOL(suid_dumpable);
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 int register_binfmt(struct linux_binfmt * fmt)
77 write_lock(&binfmt_lock);
78 list_add(&fmt->lh, &formats);
79 write_unlock(&binfmt_lock);
83 EXPORT_SYMBOL(register_binfmt);
85 void unregister_binfmt(struct linux_binfmt * fmt)
87 write_lock(&binfmt_lock);
89 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(unregister_binfmt);
94 static inline void put_binfmt(struct linux_binfmt * fmt)
96 module_put(fmt->module);
100 * Note that a shared library must be both readable and executable due to
103 * Also note that we take the address to load from from the file itself.
105 asmlinkage long sys_uselib(const char __user * library)
111 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
116 if (!S_ISREG(nd.dentry->d_inode->i_mode))
119 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
123 file = nameidata_to_filp(&nd, O_RDONLY);
124 error = PTR_ERR(file);
130 struct linux_binfmt * fmt;
132 read_lock(&binfmt_lock);
133 list_for_each_entry(fmt, &formats, lh) {
134 if (!fmt->load_shlib)
136 if (!try_module_get(fmt->module))
138 read_unlock(&binfmt_lock);
139 error = fmt->load_shlib(file);
140 read_lock(&binfmt_lock);
142 if (error != -ENOEXEC)
145 read_unlock(&binfmt_lock);
151 release_open_intent(&nd);
158 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
164 #ifdef CONFIG_STACK_GROWSUP
166 ret = expand_stack_downwards(bprm->vma, pos);
171 ret = get_user_pages(current, bprm->mm, pos,
172 1, write, 1, &page, NULL);
177 struct rlimit *rlim = current->signal->rlim;
178 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
181 * Limit to 1/4-th the stack size for the argv+env strings.
183 * - the remaining binfmt code will not run out of stack space,
184 * - the program will have a reasonable amount of stack left
187 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
196 static void put_arg_page(struct page *page)
201 static void free_arg_page(struct linux_binprm *bprm, int i)
205 static void free_arg_pages(struct linux_binprm *bprm)
209 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
212 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
215 static int __bprm_mm_init(struct linux_binprm *bprm)
218 struct vm_area_struct *vma = NULL;
219 struct mm_struct *mm = bprm->mm;
221 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
225 down_write(&mm->mmap_sem);
229 * Place the stack at the largest stack address the architecture
230 * supports. Later, we'll move this to an appropriate place. We don't
231 * use STACK_TOP because that can depend on attributes which aren't
234 vma->vm_end = STACK_TOP_MAX;
235 vma->vm_start = vma->vm_end - PAGE_SIZE;
237 vma->vm_flags = VM_STACK_FLAGS;
238 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
239 err = insert_vm_struct(mm, vma);
241 up_write(&mm->mmap_sem);
245 mm->stack_vm = mm->total_vm = 1;
246 up_write(&mm->mmap_sem);
248 bprm->p = vma->vm_end - sizeof(void *);
255 kmem_cache_free(vm_area_cachep, vma);
261 static bool valid_arg_len(struct linux_binprm *bprm, long len)
263 return len <= MAX_ARG_STRLEN;
268 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
273 page = bprm->page[pos / PAGE_SIZE];
274 if (!page && write) {
275 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
278 bprm->page[pos / PAGE_SIZE] = page;
284 static void put_arg_page(struct page *page)
288 static void free_arg_page(struct linux_binprm *bprm, int i)
291 __free_page(bprm->page[i]);
292 bprm->page[i] = NULL;
296 static void free_arg_pages(struct linux_binprm *bprm)
300 for (i = 0; i < MAX_ARG_PAGES; i++)
301 free_arg_page(bprm, i);
304 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
309 static int __bprm_mm_init(struct linux_binprm *bprm)
311 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
315 static bool valid_arg_len(struct linux_binprm *bprm, long len)
317 return len <= bprm->p;
320 #endif /* CONFIG_MMU */
323 * Create a new mm_struct and populate it with a temporary stack
324 * vm_area_struct. We don't have enough context at this point to set the stack
325 * flags, permissions, and offset, so we use temporary values. We'll update
326 * them later in setup_arg_pages().
328 int bprm_mm_init(struct linux_binprm *bprm)
331 struct mm_struct *mm = NULL;
333 bprm->mm = mm = mm_alloc();
338 err = init_new_context(current, mm);
342 err = __bprm_mm_init(bprm);
358 * count() counts the number of strings in array ARGV.
360 static int count(char __user * __user * argv, int max)
368 if (get_user(p, argv))
382 * 'copy_strings()' copies argument/environment strings from the old
383 * processes's memory to the new process's stack. The call to get_user_pages()
384 * ensures the destination page is created and not swapped out.
386 static int copy_strings(int argc, char __user * __user * argv,
387 struct linux_binprm *bprm)
389 struct page *kmapped_page = NULL;
391 unsigned long kpos = 0;
399 if (get_user(str, argv+argc) ||
400 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
405 if (!valid_arg_len(bprm, len)) {
410 /* We're going to work our way backwords. */
416 int offset, bytes_to_copy;
418 offset = pos % PAGE_SIZE;
422 bytes_to_copy = offset;
423 if (bytes_to_copy > len)
426 offset -= bytes_to_copy;
427 pos -= bytes_to_copy;
428 str -= bytes_to_copy;
429 len -= bytes_to_copy;
431 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
434 page = get_arg_page(bprm, pos, 1);
441 flush_kernel_dcache_page(kmapped_page);
442 kunmap(kmapped_page);
443 put_arg_page(kmapped_page);
446 kaddr = kmap(kmapped_page);
447 kpos = pos & PAGE_MASK;
448 flush_arg_page(bprm, kpos, kmapped_page);
450 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
459 flush_kernel_dcache_page(kmapped_page);
460 kunmap(kmapped_page);
461 put_arg_page(kmapped_page);
467 * Like copy_strings, but get argv and its values from kernel memory.
469 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
472 mm_segment_t oldfs = get_fs();
474 r = copy_strings(argc, (char __user * __user *)argv, bprm);
478 EXPORT_SYMBOL(copy_strings_kernel);
483 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
484 * the binfmt code determines where the new stack should reside, we shift it to
485 * its final location. The process proceeds as follows:
487 * 1) Use shift to calculate the new vma endpoints.
488 * 2) Extend vma to cover both the old and new ranges. This ensures the
489 * arguments passed to subsequent functions are consistent.
490 * 3) Move vma's page tables to the new range.
491 * 4) Free up any cleared pgd range.
492 * 5) Shrink the vma to cover only the new range.
494 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
496 struct mm_struct *mm = vma->vm_mm;
497 unsigned long old_start = vma->vm_start;
498 unsigned long old_end = vma->vm_end;
499 unsigned long length = old_end - old_start;
500 unsigned long new_start = old_start - shift;
501 unsigned long new_end = old_end - shift;
502 struct mmu_gather *tlb;
504 BUG_ON(new_start > new_end);
507 * ensure there are no vmas between where we want to go
510 if (vma != find_vma(mm, new_start))
514 * cover the whole range: [new_start, old_end)
516 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
519 * move the page tables downwards, on failure we rely on
520 * process cleanup to remove whatever mess we made.
522 if (length != move_page_tables(vma, old_start,
523 vma, new_start, length))
527 tlb = tlb_gather_mmu(mm, 0);
528 if (new_end > old_start) {
530 * when the old and new regions overlap clear from new_end.
532 free_pgd_range(&tlb, new_end, old_end, new_end,
533 vma->vm_next ? vma->vm_next->vm_start : 0);
536 * otherwise, clean from old_start; this is done to not touch
537 * the address space in [new_end, old_start) some architectures
538 * have constraints on va-space that make this illegal (IA64) -
539 * for the others its just a little faster.
541 free_pgd_range(&tlb, old_start, old_end, new_end,
542 vma->vm_next ? vma->vm_next->vm_start : 0);
544 tlb_finish_mmu(tlb, new_end, old_end);
547 * shrink the vma to just the new range.
549 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
554 #define EXTRA_STACK_VM_PAGES 20 /* random */
557 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
558 * the stack is optionally relocated, and some extra space is added.
560 int setup_arg_pages(struct linux_binprm *bprm,
561 unsigned long stack_top,
562 int executable_stack)
565 unsigned long stack_shift;
566 struct mm_struct *mm = current->mm;
567 struct vm_area_struct *vma = bprm->vma;
568 struct vm_area_struct *prev = NULL;
569 unsigned long vm_flags;
570 unsigned long stack_base;
572 #ifdef CONFIG_STACK_GROWSUP
573 /* Limit stack size to 1GB */
574 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
575 if (stack_base > (1 << 30))
576 stack_base = 1 << 30;
578 /* Make sure we didn't let the argument array grow too large. */
579 if (vma->vm_end - vma->vm_start > stack_base)
582 stack_base = PAGE_ALIGN(stack_top - stack_base);
584 stack_shift = vma->vm_start - stack_base;
585 mm->arg_start = bprm->p - stack_shift;
586 bprm->p = vma->vm_end - stack_shift;
588 stack_top = arch_align_stack(stack_top);
589 stack_top = PAGE_ALIGN(stack_top);
590 stack_shift = vma->vm_end - stack_top;
592 bprm->p -= stack_shift;
593 mm->arg_start = bprm->p;
597 bprm->loader -= stack_shift;
598 bprm->exec -= stack_shift;
600 down_write(&mm->mmap_sem);
601 vm_flags = vma->vm_flags;
604 * Adjust stack execute permissions; explicitly enable for
605 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
606 * (arch default) otherwise.
608 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
610 else if (executable_stack == EXSTACK_DISABLE_X)
611 vm_flags &= ~VM_EXEC;
612 vm_flags |= mm->def_flags;
614 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
620 /* Move stack pages down in memory. */
622 ret = shift_arg_pages(vma, stack_shift);
624 up_write(&mm->mmap_sem);
629 #ifdef CONFIG_STACK_GROWSUP
630 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 ret = expand_stack(vma, stack_base);
639 up_write(&mm->mmap_sem);
642 EXPORT_SYMBOL(setup_arg_pages);
644 #endif /* CONFIG_MMU */
646 struct file *open_exec(const char *name)
652 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
656 struct inode *inode = nd.dentry->d_inode;
657 file = ERR_PTR(-EACCES);
658 if (S_ISREG(inode->i_mode)) {
659 int err = vfs_permission(&nd, MAY_EXEC);
662 file = nameidata_to_filp(&nd, O_RDONLY);
664 err = deny_write_access(file);
674 release_open_intent(&nd);
680 EXPORT_SYMBOL(open_exec);
682 int kernel_read(struct file *file, unsigned long offset,
683 char *addr, unsigned long count)
691 /* The cast to a user pointer is valid due to the set_fs() */
692 result = vfs_read(file, (void __user *)addr, count, &pos);
697 EXPORT_SYMBOL(kernel_read);
699 static int exec_mmap(struct mm_struct *mm)
701 struct task_struct *tsk;
702 struct mm_struct * old_mm, *active_mm;
704 /* Notify parent that we're no longer interested in the old VM */
706 old_mm = current->mm;
707 mm_release(tsk, old_mm);
711 * Make sure that if there is a core dump in progress
712 * for the old mm, we get out and die instead of going
713 * through with the exec. We must hold mmap_sem around
714 * checking core_waiters and changing tsk->mm. The
715 * core-inducing thread will increment core_waiters for
716 * each thread whose ->mm == old_mm.
718 down_read(&old_mm->mmap_sem);
719 if (unlikely(old_mm->core_waiters)) {
720 up_read(&old_mm->mmap_sem);
725 active_mm = tsk->active_mm;
728 activate_mm(active_mm, mm);
730 arch_pick_mmap_layout(mm);
732 up_read(&old_mm->mmap_sem);
733 BUG_ON(active_mm != old_mm);
742 * This function makes sure the current process has its own signal table,
743 * so that flush_signal_handlers can later reset the handlers without
744 * disturbing other processes. (Other processes might share the signal
745 * table via the CLONE_SIGHAND option to clone().)
747 static int de_thread(struct task_struct *tsk)
749 struct signal_struct *sig = tsk->signal;
750 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
751 spinlock_t *lock = &oldsighand->siglock;
752 struct task_struct *leader = NULL;
756 * If we don't share sighandlers, then we aren't sharing anything
757 * and we can just re-use it all.
759 if (atomic_read(&oldsighand->count) <= 1) {
764 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
768 if (thread_group_empty(tsk))
769 goto no_thread_group;
772 * Kill all other threads in the thread group.
773 * We must hold tasklist_lock to call zap_other_threads.
775 read_lock(&tasklist_lock);
777 if (sig->flags & SIGNAL_GROUP_EXIT) {
779 * Another group action in progress, just
780 * return so that the signal is processed.
782 spin_unlock_irq(lock);
783 read_unlock(&tasklist_lock);
784 kmem_cache_free(sighand_cachep, newsighand);
789 * child_reaper ignores SIGKILL, change it now.
790 * Reparenting needs write_lock on tasklist_lock,
791 * so it is safe to do it under read_lock.
793 if (unlikely(tsk->group_leader == child_reaper(tsk)))
794 tsk->nsproxy->pid_ns->child_reaper = tsk;
796 zap_other_threads(tsk);
797 read_unlock(&tasklist_lock);
800 * Account for the thread group leader hanging around:
803 if (!thread_group_leader(tsk)) {
806 * The SIGALRM timer survives the exec, but needs to point
807 * at us as the new group leader now. We have a race with
808 * a timer firing now getting the old leader, so we need to
809 * synchronize with any firing (by calling del_timer_sync)
810 * before we can safely let the old group leader die.
813 spin_unlock_irq(lock);
814 if (hrtimer_cancel(&sig->real_timer))
815 hrtimer_restart(&sig->real_timer);
818 while (atomic_read(&sig->count) > count) {
819 sig->group_exit_task = tsk;
820 sig->notify_count = count;
821 __set_current_state(TASK_UNINTERRUPTIBLE);
822 spin_unlock_irq(lock);
826 sig->group_exit_task = NULL;
827 sig->notify_count = 0;
828 spin_unlock_irq(lock);
831 * At this point all other threads have exited, all we have to
832 * do is to wait for the thread group leader to become inactive,
833 * and to assume its PID:
835 if (!thread_group_leader(tsk)) {
837 * Wait for the thread group leader to be a zombie.
838 * It should already be zombie at this point, most
841 leader = tsk->group_leader;
842 while (leader->exit_state != EXIT_ZOMBIE)
846 * The only record we have of the real-time age of a
847 * process, regardless of execs it's done, is start_time.
848 * All the past CPU time is accumulated in signal_struct
849 * from sister threads now dead. But in this non-leader
850 * exec, nothing survives from the original leader thread,
851 * whose birth marks the true age of this process now.
852 * When we take on its identity by switching to its PID, we
853 * also take its birthdate (always earlier than our own).
855 tsk->start_time = leader->start_time;
857 write_lock_irq(&tasklist_lock);
859 BUG_ON(leader->tgid != tsk->tgid);
860 BUG_ON(tsk->pid == tsk->tgid);
862 * An exec() starts a new thread group with the
863 * TGID of the previous thread group. Rehash the
864 * two threads with a switched PID, and release
865 * the former thread group leader:
868 /* Become a process group leader with the old leader's pid.
869 * The old leader becomes a thread of the this thread group.
870 * Note: The old leader also uses this pid until release_task
871 * is called. Odd but simple and correct.
873 detach_pid(tsk, PIDTYPE_PID);
874 tsk->pid = leader->pid;
875 attach_pid(tsk, PIDTYPE_PID, find_pid(tsk->pid));
876 transfer_pid(leader, tsk, PIDTYPE_PGID);
877 transfer_pid(leader, tsk, PIDTYPE_SID);
878 list_replace_rcu(&leader->tasks, &tsk->tasks);
880 tsk->group_leader = tsk;
881 leader->group_leader = tsk;
883 tsk->exit_signal = SIGCHLD;
885 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
886 leader->exit_state = EXIT_DEAD;
888 write_unlock_irq(&tasklist_lock);
892 * There may be one thread left which is just exiting,
893 * but it's safe to stop telling the group to kill themselves.
900 release_task(leader);
902 if (atomic_read(&oldsighand->count) == 1) {
904 * Now that we nuked the rest of the thread group,
905 * it turns out we are not sharing sighand any more either.
906 * So we can just keep it.
908 kmem_cache_free(sighand_cachep, newsighand);
911 * Move our state over to newsighand and switch it in.
913 atomic_set(&newsighand->count, 1);
914 memcpy(newsighand->action, oldsighand->action,
915 sizeof(newsighand->action));
917 write_lock_irq(&tasklist_lock);
918 spin_lock(&oldsighand->siglock);
919 rcu_assign_pointer(tsk->sighand, newsighand);
920 spin_unlock(&oldsighand->siglock);
921 write_unlock_irq(&tasklist_lock);
923 __cleanup_sighand(oldsighand);
926 BUG_ON(!thread_group_leader(tsk));
931 * These functions flushes out all traces of the currently running executable
932 * so that a new one can be started
934 static void flush_old_files(struct files_struct * files)
939 spin_lock(&files->file_lock);
941 unsigned long set, i;
945 fdt = files_fdtable(files);
946 if (i >= fdt->max_fds)
948 set = fdt->close_on_exec->fds_bits[j];
951 fdt->close_on_exec->fds_bits[j] = 0;
952 spin_unlock(&files->file_lock);
953 for ( ; set ; i++,set >>= 1) {
958 spin_lock(&files->file_lock);
961 spin_unlock(&files->file_lock);
964 void get_task_comm(char *buf, struct task_struct *tsk)
966 /* buf must be at least sizeof(tsk->comm) in size */
968 strncpy(buf, tsk->comm, sizeof(tsk->comm));
972 void set_task_comm(struct task_struct *tsk, char *buf)
975 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
979 int flush_old_exec(struct linux_binprm * bprm)
983 struct files_struct *files;
984 char tcomm[sizeof(current->comm)];
987 * Make sure we have a private signal table and that
988 * we are unassociated from the previous thread group.
990 retval = de_thread(current);
995 * Make sure we have private file handles. Ask the
996 * fork helper to do the work for us and the exit
997 * helper to do the cleanup of the old one.
999 files = current->files; /* refcounted so safe to hold */
1000 retval = unshare_files();
1004 * Release all of the old mmap stuff
1006 retval = exec_mmap(bprm->mm);
1010 bprm->mm = NULL; /* We're using it now */
1012 /* This is the point of no return */
1013 put_files_struct(files);
1015 current->sas_ss_sp = current->sas_ss_size = 0;
1017 if (current->euid == current->uid && current->egid == current->gid)
1018 set_dumpable(current->mm, 1);
1020 set_dumpable(current->mm, suid_dumpable);
1022 name = bprm->filename;
1024 /* Copies the binary name from after last slash */
1025 for (i=0; (ch = *(name++)) != '\0';) {
1027 i = 0; /* overwrite what we wrote */
1029 if (i < (sizeof(tcomm) - 1))
1033 set_task_comm(current, tcomm);
1035 current->flags &= ~PF_RANDOMIZE;
1038 /* Set the new mm task size. We have to do that late because it may
1039 * depend on TIF_32BIT which is only updated in flush_thread() on
1040 * some architectures like powerpc
1042 current->mm->task_size = TASK_SIZE;
1044 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1046 set_dumpable(current->mm, suid_dumpable);
1047 current->pdeath_signal = 0;
1048 } else if (file_permission(bprm->file, MAY_READ) ||
1049 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1051 set_dumpable(current->mm, suid_dumpable);
1054 /* An exec changes our domain. We are no longer part of the thread
1057 current->self_exec_id++;
1059 flush_signal_handlers(current, 0);
1060 flush_old_files(current->files);
1065 reset_files_struct(current, files);
1070 EXPORT_SYMBOL(flush_old_exec);
1073 * Fill the binprm structure from the inode.
1074 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1076 int prepare_binprm(struct linux_binprm *bprm)
1079 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1082 mode = inode->i_mode;
1083 if (bprm->file->f_op == NULL)
1086 bprm->e_uid = current->euid;
1087 bprm->e_gid = current->egid;
1089 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1091 if (mode & S_ISUID) {
1092 current->personality &= ~PER_CLEAR_ON_SETID;
1093 bprm->e_uid = inode->i_uid;
1098 * If setgid is set but no group execute bit then this
1099 * is a candidate for mandatory locking, not a setgid
1102 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1103 current->personality &= ~PER_CLEAR_ON_SETID;
1104 bprm->e_gid = inode->i_gid;
1108 /* fill in binprm security blob */
1109 retval = security_bprm_set(bprm);
1113 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1114 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1117 EXPORT_SYMBOL(prepare_binprm);
1119 static int unsafe_exec(struct task_struct *p)
1122 if (p->ptrace & PT_PTRACED) {
1123 if (p->ptrace & PT_PTRACE_CAP)
1124 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1126 unsafe |= LSM_UNSAFE_PTRACE;
1128 if (atomic_read(&p->fs->count) > 1 ||
1129 atomic_read(&p->files->count) > 1 ||
1130 atomic_read(&p->sighand->count) > 1)
1131 unsafe |= LSM_UNSAFE_SHARE;
1136 void compute_creds(struct linux_binprm *bprm)
1140 if (bprm->e_uid != current->uid) {
1142 current->pdeath_signal = 0;
1147 unsafe = unsafe_exec(current);
1148 security_bprm_apply_creds(bprm, unsafe);
1149 task_unlock(current);
1150 security_bprm_post_apply_creds(bprm);
1152 EXPORT_SYMBOL(compute_creds);
1155 * Arguments are '\0' separated strings found at the location bprm->p
1156 * points to; chop off the first by relocating brpm->p to right after
1157 * the first '\0' encountered.
1159 int remove_arg_zero(struct linux_binprm *bprm)
1162 unsigned long offset;
1170 offset = bprm->p & ~PAGE_MASK;
1171 page = get_arg_page(bprm, bprm->p, 0);
1176 kaddr = kmap_atomic(page, KM_USER0);
1178 for (; offset < PAGE_SIZE && kaddr[offset];
1179 offset++, bprm->p++)
1182 kunmap_atomic(kaddr, KM_USER0);
1185 if (offset == PAGE_SIZE)
1186 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1187 } while (offset == PAGE_SIZE);
1196 EXPORT_SYMBOL(remove_arg_zero);
1199 * cycle the list of binary formats handler, until one recognizes the image
1201 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1204 struct linux_binfmt *fmt;
1206 /* handle /sbin/loader.. */
1208 struct exec * eh = (struct exec *) bprm->buf;
1210 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1211 (eh->fh.f_flags & 0x3000) == 0x3000)
1214 unsigned long loader;
1216 allow_write_access(bprm->file);
1220 loader = bprm->vma->vm_end - sizeof(void *);
1222 file = open_exec("/sbin/loader");
1223 retval = PTR_ERR(file);
1227 /* Remember if the application is TASO. */
1228 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1231 bprm->loader = loader;
1232 retval = prepare_binprm(bprm);
1235 /* should call search_binary_handler recursively here,
1236 but it does not matter */
1240 retval = security_bprm_check(bprm);
1244 /* kernel module loader fixup */
1245 /* so we don't try to load run modprobe in kernel space. */
1248 retval = audit_bprm(bprm);
1253 for (try=0; try<2; try++) {
1254 read_lock(&binfmt_lock);
1255 list_for_each_entry(fmt, &formats, lh) {
1256 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1259 if (!try_module_get(fmt->module))
1261 read_unlock(&binfmt_lock);
1262 retval = fn(bprm, regs);
1265 allow_write_access(bprm->file);
1269 current->did_exec = 1;
1270 proc_exec_connector(current);
1273 read_lock(&binfmt_lock);
1275 if (retval != -ENOEXEC || bprm->mm == NULL)
1278 read_unlock(&binfmt_lock);
1282 read_unlock(&binfmt_lock);
1283 if (retval != -ENOEXEC || bprm->mm == NULL) {
1287 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1288 if (printable(bprm->buf[0]) &&
1289 printable(bprm->buf[1]) &&
1290 printable(bprm->buf[2]) &&
1291 printable(bprm->buf[3]))
1292 break; /* -ENOEXEC */
1293 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1300 EXPORT_SYMBOL(search_binary_handler);
1303 * sys_execve() executes a new program.
1305 int do_execve(char * filename,
1306 char __user *__user *argv,
1307 char __user *__user *envp,
1308 struct pt_regs * regs)
1310 struct linux_binprm *bprm;
1312 unsigned long env_p;
1316 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1320 file = open_exec(filename);
1321 retval = PTR_ERR(file);
1328 bprm->filename = filename;
1329 bprm->interp = filename;
1331 retval = bprm_mm_init(bprm);
1335 bprm->argc = count(argv, MAX_ARG_STRINGS);
1336 if ((retval = bprm->argc) < 0)
1339 bprm->envc = count(envp, MAX_ARG_STRINGS);
1340 if ((retval = bprm->envc) < 0)
1343 retval = security_bprm_alloc(bprm);
1347 retval = prepare_binprm(bprm);
1351 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1355 bprm->exec = bprm->p;
1356 retval = copy_strings(bprm->envc, envp, bprm);
1361 retval = copy_strings(bprm->argc, argv, bprm);
1364 bprm->argv_len = env_p - bprm->p;
1366 retval = search_binary_handler(bprm,regs);
1368 /* execve success */
1369 free_arg_pages(bprm);
1370 security_bprm_free(bprm);
1371 acct_update_integrals(current);
1377 free_arg_pages(bprm);
1379 security_bprm_free(bprm);
1387 allow_write_access(bprm->file);
1397 int set_binfmt(struct linux_binfmt *new)
1399 struct linux_binfmt *old = current->binfmt;
1402 if (!try_module_get(new->module))
1405 current->binfmt = new;
1407 module_put(old->module);
1411 EXPORT_SYMBOL(set_binfmt);
1413 /* format_corename will inspect the pattern parameter, and output a
1414 * name into corename, which must have space for at least
1415 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1417 static int format_corename(char *corename, const char *pattern, long signr)
1419 const char *pat_ptr = pattern;
1420 char *out_ptr = corename;
1421 char *const out_end = corename + CORENAME_MAX_SIZE;
1423 int pid_in_pattern = 0;
1426 if (*pattern == '|')
1429 /* Repeat as long as we have more pattern to process and more output
1432 if (*pat_ptr != '%') {
1433 if (out_ptr == out_end)
1435 *out_ptr++ = *pat_ptr++;
1437 switch (*++pat_ptr) {
1440 /* Double percent, output one percent */
1442 if (out_ptr == out_end)
1449 rc = snprintf(out_ptr, out_end - out_ptr,
1450 "%d", current->tgid);
1451 if (rc > out_end - out_ptr)
1457 rc = snprintf(out_ptr, out_end - out_ptr,
1458 "%d", current->uid);
1459 if (rc > out_end - out_ptr)
1465 rc = snprintf(out_ptr, out_end - out_ptr,
1466 "%d", current->gid);
1467 if (rc > out_end - out_ptr)
1471 /* signal that caused the coredump */
1473 rc = snprintf(out_ptr, out_end - out_ptr,
1475 if (rc > out_end - out_ptr)
1479 /* UNIX time of coredump */
1482 do_gettimeofday(&tv);
1483 rc = snprintf(out_ptr, out_end - out_ptr,
1485 if (rc > out_end - out_ptr)
1492 down_read(&uts_sem);
1493 rc = snprintf(out_ptr, out_end - out_ptr,
1494 "%s", utsname()->nodename);
1496 if (rc > out_end - out_ptr)
1502 rc = snprintf(out_ptr, out_end - out_ptr,
1503 "%s", current->comm);
1504 if (rc > out_end - out_ptr)
1508 /* core limit size */
1510 rc = snprintf(out_ptr, out_end - out_ptr,
1511 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1512 if (rc > out_end - out_ptr)
1522 /* Backward compatibility with core_uses_pid:
1524 * If core_pattern does not include a %p (as is the default)
1525 * and core_uses_pid is set, then .%pid will be appended to
1526 * the filename. Do not do this for piped commands. */
1527 if (!ispipe && !pid_in_pattern
1528 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1529 rc = snprintf(out_ptr, out_end - out_ptr,
1530 ".%d", current->tgid);
1531 if (rc > out_end - out_ptr)
1540 static void zap_process(struct task_struct *start)
1542 struct task_struct *t;
1544 start->signal->flags = SIGNAL_GROUP_EXIT;
1545 start->signal->group_stop_count = 0;
1549 if (t != current && t->mm) {
1550 t->mm->core_waiters++;
1551 sigaddset(&t->pending.signal, SIGKILL);
1552 signal_wake_up(t, 1);
1554 } while ((t = next_thread(t)) != start);
1557 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1560 struct task_struct *g, *p;
1561 unsigned long flags;
1564 spin_lock_irq(&tsk->sighand->siglock);
1565 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1566 tsk->signal->group_exit_code = exit_code;
1570 spin_unlock_irq(&tsk->sighand->siglock);
1574 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1578 for_each_process(g) {
1579 if (g == tsk->group_leader)
1587 * p->sighand can't disappear, but
1588 * may be changed by de_thread()
1590 lock_task_sighand(p, &flags);
1592 unlock_task_sighand(p, &flags);
1596 } while ((p = next_thread(p)) != g);
1600 return mm->core_waiters;
1603 static int coredump_wait(int exit_code)
1605 struct task_struct *tsk = current;
1606 struct mm_struct *mm = tsk->mm;
1607 struct completion startup_done;
1608 struct completion *vfork_done;
1611 init_completion(&mm->core_done);
1612 init_completion(&startup_done);
1613 mm->core_startup_done = &startup_done;
1615 core_waiters = zap_threads(tsk, mm, exit_code);
1616 up_write(&mm->mmap_sem);
1618 if (unlikely(core_waiters < 0))
1622 * Make sure nobody is waiting for us to release the VM,
1623 * otherwise we can deadlock when we wait on each other
1625 vfork_done = tsk->vfork_done;
1627 tsk->vfork_done = NULL;
1628 complete(vfork_done);
1632 wait_for_completion(&startup_done);
1634 BUG_ON(mm->core_waiters);
1635 return core_waiters;
1639 * set_dumpable converts traditional three-value dumpable to two flags and
1640 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1641 * these bits are not changed atomically. So get_dumpable can observe the
1642 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1643 * return either old dumpable or new one by paying attention to the order of
1644 * modifying the bits.
1646 * dumpable | mm->flags (binary)
1647 * old new | initial interim final
1648 * ---------+-----------------------
1656 * (*) get_dumpable regards interim value of 10 as 11.
1658 void set_dumpable(struct mm_struct *mm, int value)
1662 clear_bit(MMF_DUMPABLE, &mm->flags);
1664 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1667 set_bit(MMF_DUMPABLE, &mm->flags);
1669 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1672 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1674 set_bit(MMF_DUMPABLE, &mm->flags);
1678 EXPORT_SYMBOL_GPL(set_dumpable);
1680 int get_dumpable(struct mm_struct *mm)
1684 ret = mm->flags & 0x3;
1685 return (ret >= 2) ? 2 : ret;
1688 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1690 char corename[CORENAME_MAX_SIZE + 1];
1691 struct mm_struct *mm = current->mm;
1692 struct linux_binfmt * binfmt;
1693 struct inode * inode;
1696 int fsuid = current->fsuid;
1699 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1700 char **helper_argv = NULL;
1701 int helper_argc = 0;
1704 audit_core_dumps(signr);
1706 binfmt = current->binfmt;
1707 if (!binfmt || !binfmt->core_dump)
1709 down_write(&mm->mmap_sem);
1710 if (!get_dumpable(mm)) {
1711 up_write(&mm->mmap_sem);
1716 * We cannot trust fsuid as being the "true" uid of the
1717 * process nor do we know its entire history. We only know it
1718 * was tainted so we dump it as root in mode 2.
1720 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1721 flag = O_EXCL; /* Stop rewrite attacks */
1722 current->fsuid = 0; /* Dump root private */
1724 set_dumpable(mm, 0);
1726 retval = coredump_wait(exit_code);
1731 * Clear any false indication of pending signals that might
1732 * be seen by the filesystem code called to write the core file.
1734 clear_thread_flag(TIF_SIGPENDING);
1737 * lock_kernel() because format_corename() is controlled by sysctl, which
1738 * uses lock_kernel()
1741 ispipe = format_corename(corename, core_pattern, signr);
1744 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1745 * to a pipe. Since we're not writing directly to the filesystem
1746 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1747 * created unless the pipe reader choses to write out the core file
1748 * at which point file size limits and permissions will be imposed
1749 * as it does with any other process
1751 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1755 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1756 /* Terminate the string before the first option */
1757 delimit = strchr(corename, ' ');
1760 delimit = strrchr(helper_argv[0], '/');
1764 delimit = helper_argv[0];
1765 if (!strcmp(delimit, current->comm)) {
1766 printk(KERN_NOTICE "Recursive core dump detected, "
1771 core_limit = RLIM_INFINITY;
1773 /* SIGPIPE can happen, but it's just never processed */
1774 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1776 printk(KERN_INFO "Core dump to %s pipe failed\n",
1781 file = filp_open(corename,
1782 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1786 inode = file->f_path.dentry->d_inode;
1787 if (inode->i_nlink > 1)
1788 goto close_fail; /* multiple links - don't dump */
1789 if (!ispipe && d_unhashed(file->f_path.dentry))
1792 /* AK: actually i see no reason to not allow this for named pipes etc.,
1793 but keep the previous behaviour for now. */
1794 if (!ispipe && !S_ISREG(inode->i_mode))
1798 if (!file->f_op->write)
1800 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1803 retval = binfmt->core_dump(signr, regs, file, core_limit);
1806 current->signal->group_exit_code |= 0x80;
1808 filp_close(file, NULL);
1811 argv_free(helper_argv);
1813 current->fsuid = fsuid;
1814 complete_all(&mm->core_done);